Compressing inverted indexes with recursive graph bisection: A reproducibility study

Document reordering is an important but often overlooked preprocessing stage in index construction. Reordering document identifiers in graphs and inverted indexes has been shown to reduce storage costs and improve processing efficiency in the resulting indexes. However, surprisingly few document reordering algorithms are publicly available despite their importance. A new reordering algorithm derived from recursive graph bisection was recently proposed by Dhulipala et al., and shown to be highly effective and efficient when compared against other state-of-the-art reordering strategies. In this work, we present a reproducibility study of this new algorithm. We describe the implementation challenges encountered, and explore the performance characteristics of our clean-room reimplementation. We show that we are able to successfully reproduce the core results of the original paper, and show that the algorithm generalizes to other collections and indexing frameworks. Furthermore, we make our implementation publicly available to help promote further research in this space.